Software Engineer's Guide to AI Agents is the practical handbook for developers who need to move past demos and into production fast. Written in a conversational style with real-world analogies, this book is designed to be picked up and read cover to cover in a weekend. It offers clear, concise explanations and Python-based examples that get straight to the point.
Written by Harish Dasari, a software engineer at Flatiron Health building agentic applications, this book takes you from the fundamentals of LLMs, such as tokens, context windows, and prompt engineering, all the way to deploying, securing, and scaling autonomous AI systems.
You will learn how to give agents tools and memory using Pydantic and FastAPI, orchestrate complex workflows with state machines and graphs, and connect to any system using the Model Context Protocol (MCP). You will also master the production engineering that separates prototypes from products: observability with OpenTelemetry, LLM-as-a-Judge evaluation, defense-in-depth security, cost optimization, and human-in-the-loop patterns for high-stakes environments.
What you will learn:
- How LLMs actually work and how to treat them as software components, not magic
- Tool use, RAG, and structured outputs with Pydantic
- Agent orchestration patterns: ReAct, multi-agent systems, and the Supervisor vs. Network tradeoff
- Production deployment with FastAPI, async patterns, and sandboxed code execution
- Observability, evaluation, and security for non-deterministic systems
- Cost optimization, caching strategies, and model routing
- Integrating with legacy business systems and the future of multimodal agents
A quick, accessible read written for Python developers but readable by anyone who builds software.
